Part 8/12:

He also delved into techniques for model updates. Since satellites are remote and hardware-constrained, updating models mid-mission is challenging but can be approached through ground station commands or more advanced on-orbit learning methods—an area still ripe for innovation.

Overcoming Hardware and System Constraints

A significant part of Akash’s talk focused on managing hardware limitations:

• GPUs like Nvidia H100 are power-hungry and generate excess heat—unsuitable for space.

• The industry is increasingly adopting FPGA processors, which are more energy-efficient and radiation-tolerant but currently lag in deep learning capabilities.

• Practical AI deployment involves carefully balancing model complexity, power use, and hardware robustness.